14. Image Processing Techniques

Many composites in Geo2Grid take advantage of various corrections or adjustments to produce the best looking imagery possible. The below sections describe the corrections and other related topics used in Geo2Grid. See the various Readers documentation for more information on what products are available and descriptions of what corrections are used.

14.1. RGB Images

Satellite imagers can simultaneously observe the Earth in multiple spectral channels, while the human eye is sensitive to only the visible channels. By mapping data from imager channels to the visible red, green, and blue channels in different ways, we can produce “RGB” images that show the Earth as a human would see it from space (“true color”), or that emphasize certain features that can be detected using combinations of different channels (“false color”).

Luminance (L), or single band, images are also used when displaying a single imager channel in grayscale. Another popular way of showing single imager channels is to apply a “colormap” to the data. In these cases, each data value of a single satellite imager channel is represented by a color. This is different than the RGB composites described above where multiple channels go into making a single color image.

Depending on the configuration and writer used, Geo2Grid may also add an additional “Alpha” channel (ex. RGBA) to an image. This Alpha channel is used to determine the opaqueness or transparency of an image. This is typically used in Geo2Grid to make invalid or missing data values transparent (completely opaque or completely transparent).

14.2. Solar Zenith Angle Modification

Reflectance is defined as the reflected radiation as a fraction of the incident radiation. To calculate reflectance, the solar zenith angle is needed, in addition to the radiance measured by the sensor. This modification, used by some RGB recipes, involves dividing the channel data by the cosine of the solar zenith angle.

14.3. Rayleigh Scattering Correction - CREFL

Due to the size of molecules that make up our atmosphere, some visible channel light is preferentially scattered more than others, especially at larger viewing angles. The Corrected Reflectance algorithm performs a simple atmospheric correction with MODIS visible, near-infrared, and short-wave infrared bands (1 to 16). Later versions of the software were adapted to work with VIIRS data. Both implementations have been merged and made available as a “modifier” in the Satpy Python library and used by Geo2Grid. This algorithm was originally developed by the MODIS Rapid Response Team (http://rapidfire.sci.gsfc.nasa.gov/) and made available by cooperative agreement, with subsequent additions by the University of South Florida (USF) and the NASA Direct Readout Laboratory (DRL).

The algorithm corrects for molecular (Rayleigh) scattering and gaseous absorption (water vapor, ozone) using climatological values for gas contents. It requires no real-time input of ancillary data. The algorithm performs no aerosol correction. The Corrected Reflectance products are very similar to the MODIS Land Surface Reflectance product (MOD09) in clear atmospheric conditions, since the algorithms used to derive both are based on the 6S Radiative Transfer Model (Vermote et al.1994). The products show differences in the presence of aerosols, however, because the MODIS Land Surface Reflectance product uses a more complex atmospheric correction algorithm that includes a correction for aerosols.

14.4. Rayleigh Scattering Correction - Pyspectral

Due to the size of molecules that make up our atmosphere, some visible channel light is preferentially scattered more than others, especially at larger viewing angles. One method to correct for this is implemented in the Pyspectral Python library. A detailed description of the algorithm used by Pyspectral and other features of the library can be found in the official Pyspectral documentation:

https://pyspectral.readthedocs.io/en/latest/rayleigh_correction.html

14.5. Ratio Sharpening

Some sensors include channels that measure radiance at the same wavelength, but at different spatial resolutions. When making an RGB image that uses one of these multi-resolution wavelengths combined with other channels that are only available at lower resolutions, we can use the multi-resolution channels to sharpen the other channels. For example, if the high-resolution channel is used for R, and lower resolution channels for G and B, we can do:

R_ratio = R_hi / R_lo
new_R = R_hi
new_G = G * R_ratio
new_B = B * R_ratio

By upsampling the lower resolution G and B channels and multiplying by the ratio of high and low resolution R channels, we can produce a sharper looking final image. That is, the lower resolution channels appear to have a better spatial resolution than they did originally.

14.6. Self Ratio Sharpening

Similar to the Ratio Sharpening described above, it is possible to apply a similar sharpening when one of the channels of the RGB is only provided in a high resolution. In this case, we can downsample the high resolution channel to the resolution of the other channels (averaging the pixels), then upsample the result again. By taking the ratio of the original high resolution and this averaged version, we can produce a ratio similar to that in the above ratio sharpening technique.

14.7. Non-linear True Color Scaling

As a final step for some RGB images, Geo2Grid scales the image values using a series of linear interpolation ranges to bring out certain regions of the image and lessen the effect of others. For lack of a better name, these multiple linear stretches make up an overall non-linear scaling. A typical scaling where reflectance data (0 - 1) has been multiplied by 255 (8-bit unsigned integer) would be:

Input Range	Output Range
0 - 25	0 - 90
25 - 55	90 - 140
55 - 100	140 - 175
100 - 255	175 - 255